nshtrainer 0.10.10__py3-none-any.whl → 0.10.12__py3-none-any.whl

nshtrainer/_checkpoint/metadata.py

@@ -1,6 +1,8 @@
 import copy
 import datetime
 import logging
+import shutil
+from collections.abc import Callable
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, cast
 
@@ -8,7 +10,7 @@ import nshconfig as C
 import numpy as np
 import torch
 
-from ..model._environment import EnvironmentConfig
+from ..util._environment_info import EnvironmentConfig
 
 if TYPE_CHECKING:
     from ..model import BaseConfig, LightningModuleBase
@@ -100,3 +102,74 @@ def _write_checkpoint_metadata(
         log.warning(f"Failed to write hparams to {checkpoint_path}: {e}")
     else:
         log.info(f"Checkpoint metadata written to {checkpoint_path}")
+
+
+def _remove_checkpoint_metadata(checkpoint_path: Path):
+    for path in (
+        checkpoint_path.with_suffix(METADATA_PATH_SUFFIX),
+        checkpoint_path.with_suffix(HPARAMS_PATH_SUFFIX),
+    ):
+        try:
+            path.unlink(missing_ok=True)
+        except Exception as e:
+            log.warning(f"Failed to remove {path}: {e}")
+        else:
+            log.info(f"Removed {path}")
+
+
+def _link_checkpoint_metadata(checkpoint_path: Path, linked_checkpoint_path: Path):
+    # First, remove any existing metadata files
+    _remove_checkpoint_metadata(linked_checkpoint_path)
+
+    # Link the metadata files to the new checkpoint
+    for path in (
+        checkpoint_path.with_suffix(METADATA_PATH_SUFFIX),
+        checkpoint_path.with_suffix(HPARAMS_PATH_SUFFIX),
+    ):
+        linked_path = linked_checkpoint_path.with_suffix(path.suffix)
+        try:
+            try:
+                linked_path.symlink_to(path)
+            except OSError:
+                # on Windows, special permissions are required to create symbolic links as a regular user
+                # fall back to copying the file
+                shutil.copy(path, linked_path)
+        except Exception as e:
+            log.warning(f"Failed to link {path} to {linked_path}: {e}")
+        else:
+            log.info(f"Linked {path} to {linked_path}")
+
+
+def _checkpoint_sort_key_fn(key: Callable[[CheckpointMetadata, Path], Any]):
+    def sort_key_fn(checkpoint_path: Path):
+        if not (p := checkpoint_path.with_suffix(METADATA_PATH_SUFFIX)).exists():
+            raise FileNotFoundError(f"Metadata file not found: {p}")
+
+        nonlocal key
+        return key(CheckpointMetadata.from_file(p), p)
+
+    return sort_key_fn
+
+
+def _sort_ckpts_by_metadata(
+    checkpoint_paths: list[Path],
+    key: Callable[[CheckpointMetadata, Path], Any],
+    fallback_key: Callable[[Path], Any],
+):
+    # First, let's make sure all the metadata files exist.
+    # If not, use the fallback function to sort the checkpoints.
+    no_metadata_paths: list[Path] = []
+    for path in checkpoint_paths:
+        if (path.with_suffix(METADATA_PATH_SUFFIX)).exists():
+            continue
+
+        no_metadata_paths.append(path)
+
+    if no_metadata_paths:
+        log.warning(
+            f"Metadata file not found on {len(no_metadata_paths)} checkpoints: {no_metadata_paths}\n"
+            "Falling back to sorting by last modified time."
+        )
+        return sorted(checkpoint_paths, key=fallback_key)
+
+    return sorted(checkpoint_paths, key=_checkpoint_sort_key_fn(key))

nshtrainer/_checkpoint/saver.py

@@ -0,0 +1,52 @@
+import os
+import shutil
+from pathlib import Path
+
+from lightning.pytorch import Trainer
+
+from .metadata import _link_checkpoint_metadata, _remove_checkpoint_metadata
+
+
+def _link_checkpoint(
+    trainer: Trainer,
+    filepath: str | Path | os.PathLike,
+    linkpath: str | Path | os.PathLike,
+    *,
+    barrier: bool,
+    metadata: bool,
+):
+    if not isinstance(filepath, Path):
+        filepath = Path(filepath)
+    if not isinstance(linkpath, Path):
+        linkpath = Path(linkpath)
+
+    if trainer.is_global_zero:
+        if linkpath.exists():
+            if linkpath.is_symlink() or linkpath.is_file():
+                linkpath.unlink()
+            elif linkpath.is_dir():
+                shutil.rmtree(linkpath)
+            _remove_checkpoint_metadata(linkpath)
+
+        try:
+            target_path = filepath.relative_to(linkpath.parent)
+            linkpath.symlink_to(target_path)
+        except OSError:
+            # on Windows, special permissions are required to create symbolic links as a regular user
+            # fall back to copying the file
+            shutil.copy(filepath, linkpath)
+
+        _link_checkpoint_metadata(filepath, linkpath)
+    if barrier:
+        trainer.strategy.barrier()
+
+
+def _remove_checkpoint(
+    trainer: Trainer,
+    filepath: str | Path | os.PathLike,
+    remove_metadata: bool = True,
+):
+    if not isinstance(filepath, Path):
+        filepath = Path(filepath)
+    trainer.strategy.remove_checkpoint(filepath)
+    _remove_checkpoint_metadata(filepath)

nshtrainer/callbacks/latest_epoch_checkpoint.py

@@ -6,6 +6,8 @@ from lightning.pytorch import LightningModule, Trainer
 from lightning.pytorch.callbacks import Checkpoint
 from typing_extensions import override
 
+from .._checkpoint.metadata import _sort_ckpts_by_metadata
+from .._checkpoint.saver import _link_checkpoint, _remove_checkpoint
 from .base import CallbackConfigBase
 
 log = logging.getLogger(__name__)
@@ -17,15 +19,18 @@ class LatestEpochCheckpointCallbackConfig(CallbackConfigBase):
     dirpath: str | Path | None = None
     """Directory path to save the checkpoint file."""
 
-    filename: str = "latest_epoch{epoch:02d}_step{step:04d}.ckpt"
+    filename: str = "epoch{epoch:02d}_step{step:04d}"
     """Checkpoint filename. This must not include the extension."""
 
     save_weights_only: bool = False
     """Whether to save only the model's weights or the entire model object."""
 
-    latest_symlink_filename: str | None = "latest.ckpt"
+    latest_symlink_filename: str | None = "latest"
     """Filename for the latest symlink. If None, no symlink will be created."""
 
+    latest_k: int | Literal["all"] = 1
+    """Number of latest checkpoints to keep. If "all", all checkpoints are kept."""
+
     @override
     def create_callbacks(self, root_config):
         dirpath = self.dirpath or root_config.directory.resolve_subdirectory(
@@ -37,38 +42,73 @@ class LatestEpochCheckpointCallbackConfig(CallbackConfigBase):
 
 
 class LatestEpochCheckpoint(Checkpoint):
+    PREFIX = "latest_"
+    EXTENSION = ".ckpt"
+
     def __init__(self, config: LatestEpochCheckpointCallbackConfig, dirpath: Path):
         super().__init__()
 
         self.config = config
         self.dirpath = dirpath
 
+    @override
+    def on_train_epoch_end(self, trainer: Trainer, pl_module: LightningModule):
+        self._save_new_checkpoint(trainer)
+
+    def _latest_symlink_filename(self):
+        if (filename := self.config.latest_symlink_filename) is None:
+            return None
+        return f"{filename}{self.EXTENSION}"
+
     def _ckpt_path(self, trainer: Trainer):
-        return self.dirpath / self.config.filename.format(
+        filename = self.config.filename.format(
             epoch=trainer.current_epoch, step=trainer.global_step
         )
+        filename = f"{self.PREFIX}{filename}.{self.EXTENSION}"
+        return self.dirpath / filename
+
+    def _remove_checkpoints(self, trainer: Trainer, ckpt_paths: list[Path]):
+        for ckpt_path in ckpt_paths:
+            _remove_checkpoint(trainer, ckpt_path, remove_metadata=True)
+
+    def _remove_old_checkpoints(self, trainer: Trainer):
+        if (latest_k := self.config.latest_k) == "all":
+            return
+
+        # Get all configs, ignoring the latest symlink
+        ckpt_paths = list(self.dirpath.glob(f"{self.PREFIX}*{self.EXTENSION}"))
+        # Ignore the latest symlink
+        if (latest_symlink_filename := self._latest_symlink_filename()) is not None:
+            ckpt_paths = [p for p in ckpt_paths if p.name != latest_symlink_filename]
+
+        # Sort by epoch, then step, then last modified
+        ckpt_paths = _sort_ckpts_by_metadata(
+            ckpt_paths,
+            key=lambda meta, p: (meta.epoch, meta.global_step, p.stat().st_mtime),
+            fallback_key=lambda p: p.stat().st_mtime,
+            # ^ Called if metadata is not found on all checkpoints
+        )
+
+        # Remove all but the latest k checkpoints
+        ckpts_to_remove = ckpt_paths[:-latest_k]
+        self._remove_checkpoints(trainer, ckpts_to_remove)
+
+    def _save_new_checkpoint(self, trainer: Trainer):
+        # Remove old checkpoints
+        self._remove_old_checkpoints(trainer)
 
-    @override
-    def on_train_epoch_end(self, trainer: Trainer, pl_module: LightningModule):
         # Save the new checkpoint
         filepath = self._ckpt_path(trainer)
         trainer.save_checkpoint(filepath, self.config.save_weights_only)
 
         # Create the latest symlink
-        if (
-            trainer.is_global_zero
-            and (symlink_filename := self.config.latest_symlink_filename) is not None
-        ):
+        if (symlink_filename := self._latest_symlink_filename()) is not None:
             symlink_path = self.dirpath / symlink_filename
-            symlink_path.unlink(missing_ok=True)
-            symlink_path.symlink_to(filepath.name)
+            _link_checkpoint(
+                trainer,
+                filepath,
+                symlink_path,
+                barrier=True,
+                metadata=True,
+            )
             log.info(f"Created latest symlink: {symlink_path}")
-
-    def latest_checkpoint(self):
-        if (symlink_filename := self.config.latest_symlink_filename) is None:
-            return None
-
-        if not (symlink_path := self.dirpath / symlink_filename).exists():
-            return None
-
-        return symlink_path

nshtrainer/callbacks/model_checkpoint.py

@@ -4,11 +4,13 @@ from datetime import timedelta
 from pathlib import Path
 from typing import TYPE_CHECKING, Literal
 
+from lightning.pytorch import Trainer
 from lightning.pytorch.callbacks.model_checkpoint import (
     ModelCheckpoint as _ModelCheckpoint,
 )
 from typing_extensions import override
 
+from .._checkpoint.saver import _link_checkpoint, _remove_checkpoint
 from ..metrics import MetricConfig
 from .base import CallbackConfigBase
 
@@ -158,6 +160,8 @@ class ModelCheckpointCallbackConfig(CallbackConfigBase):
 
 
 class ModelCheckpoint(_ModelCheckpoint):
+    CHECKPOINT_NAME_LAST = "best"
+
     @override
     def __init__(
         self,
@@ -185,3 +189,17 @@ class ModelCheckpoint(_ModelCheckpoint):
             save_on_train_epoch_end=self.config.save_on_train_epoch_end,
             enable_version_counter=self.config.enable_version_counter,
         )
+
+    @override
+    def _link_checkpoint(self, trainer: Trainer, filepath: str, linkpath: str):  # pyright: ignore[reportIncompatibleMethodOverride]
+        return _link_checkpoint(
+            trainer,
+            filepath,
+            linkpath,
+            barrier=True,
+            metadata=True,
+        )
+
+    @override
+    def _remove_checkpoint(self, trainer: Trainer, filepath: str):
+        return _remove_checkpoint(trainer, filepath, remove_metadata=True)

nshtrainer/ll/model.py

@@ -1 +1,12 @@
 from nshtrainer.model import *  # noqa: F403
+
+from ..util._environment_info import (
+    EnvironmentClassInformationConfig as EnvironmentClassInformationConfig,
+)
+from ..util._environment_info import EnvironmentConfig as EnvironmentConfig
+from ..util._environment_info import (
+    EnvironmentLinuxEnvironmentConfig as EnvironmentLinuxEnvironmentConfig,
+)
+from ..util._environment_info import (
+    EnvironmentSLURMInformationConfig as EnvironmentSLURMInformationConfig,
+)

nshtrainer/model/__init__.py

@@ -1,16 +1,5 @@
 from typing_extensions import TypeAlias
 
-from ._environment import (
-    EnvironmentClassInformationConfig as EnvironmentClassInformationConfig,
-)
-from ._environment import EnvironmentConfig as EnvironmentConfig
-from ._environment import (
-    EnvironmentLinuxEnvironmentConfig as EnvironmentLinuxEnvironmentConfig,
-)
-from ._environment import (
-    EnvironmentSLURMInformationConfig as EnvironmentSLURMInformationConfig,
-)
-from ._environment import EnvironmentSnapshotConfig as EnvironmentSnapshotConfig
 from .base import Base as Base
 from .base import LightningModuleBase as LightningModuleBase
 from .config import BaseConfig as BaseConfig

nshtrainer/model/base.py

@@ -11,7 +11,7 @@ from lightning.pytorch.callbacks import Callback
 from lightning.pytorch.utilities.types import STEP_OUTPUT
 from typing_extensions import Self, TypeVar, override
 
-from ._environment import EnvironmentConfig
+from ..util._environment_info import EnvironmentConfig
 from .config import BaseConfig
 from .modules.callback import CallbackModuleMixin
 from .modules.debug import DebugModuleMixin

nshtrainer/model/config.py

@@ -44,7 +44,7 @@ from ..callbacks import (
 )
 from ..callbacks.base import CallbackConfigBase
 from ..metrics import MetricConfig
-from ._environment import EnvironmentConfig
+from ..util._environment_info import EnvironmentConfig
 
 log = logging.getLogger(__name__)
 

nshtrainer/{model/_environment.py → util/_environment_info.py}

@@ -14,11 +14,11 @@ import psutil
 import torch
 from typing_extensions import Self
 
-from ..util.slurm import parse_slurm_node_list
+from .slurm import parse_slurm_node_list
 
 if TYPE_CHECKING:
-    from .base import LightningModuleBase
-    from .config import BaseConfig
+    from ..model.base import LightningModuleBase
+    from ..model.config import BaseConfig
 
 
 log = logging.getLogger(__name__)

{nshtrainer-0.10.10.dist-info → nshtrainer-0.10.12.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nshtrainer
-Version: 0.10.10
+Version: 0.10.12
 Summary: 
 Author: Nima Shoghi
 Author-email: nimashoghi@gmail.com

{nshtrainer-0.10.10.dist-info → nshtrainer-0.10.12.dist-info}/RECORD

@@ -1,6 +1,7 @@
 nshtrainer/__init__.py,sha256=39loiLLXbaGiozEsAn8mPHopxaPsek8JsgR9DD2gxtY,583
 nshtrainer/_checkpoint/loader.py,sha256=48flPr1XgQHOgIPaCrRqOEvRuG0SZuV3cQ1vgHLqFqI,11025
-nshtrainer/_checkpoint/metadata.py,sha256=C7je_soYyEbZjiq7p2_pSVFkgcXnz2J2H5sMy8oskx0,3051
+nshtrainer/_checkpoint/metadata.py,sha256=soK9tXVs6EOpzhlnIxTEF51KmdkaCDUj0Rdyid3uREk,5640
+nshtrainer/_checkpoint/saver.py,sha256=KZp9ITUVHwj2Ttu81zXKdlS_h-fKkHearspwuAijDpM,1501
 nshtrainer/_experimental/__init__.py,sha256=2tQIcrWT8U8no_AeBTYnozaTmxN40kuAJdGQ4b-PoWM,120
 nshtrainer/_experimental/flops/__init__.py,sha256=edo9Ez3LlrnxkNRX9W6YBhPkRPKYGLpkpnl5gx7sEX8,1550
 nshtrainer/_experimental/flops/flop_counter.py,sha256=-sL0Fy6poXa__hyzUMdZScjPULp4coQELQpPU6p6dXU,25736
@@ -14,9 +15,9 @@ nshtrainer/callbacks/ema.py,sha256=8-WHmKFP3VfnzMviJaIFmVD9xHPqIPmq9NRF5xdu3c8,1
 nshtrainer/callbacks/finite_checks.py,sha256=gJC_RUr3ais3FJI0uB6wUZnDdE3WRwCix3ppA3PwQXA,2077
 nshtrainer/callbacks/gradient_skipping.py,sha256=pqu5AELx4ctJxR2Y7YSSiGd5oGauVCTZFCEIIS6s88w,3665
 nshtrainer/callbacks/interval.py,sha256=smz5Zl8cN6X6yHKVsMRS2e3SEkzRCP3LvwE1ONvLfaw,8080
-nshtrainer/callbacks/latest_epoch_checkpoint.py,sha256=zCRAUsqW-2PaoIwVKlXOqdh2uF_B_YUUTmQO1wSomR8,2489
+nshtrainer/callbacks/latest_epoch_checkpoint.py,sha256=t4vWa4PvJDO3rKXKZbuegm7iLl7xCEd17wNif0Bp-BA,4138
 nshtrainer/callbacks/log_epoch.py,sha256=fTa_K_Y8A7g09630cG4YkDE6AzSMPkjb9bpPm4gtqos,1120
-nshtrainer/callbacks/model_checkpoint.py,sha256=wkT8sHGkIuatm4gSn4W-fqicA_HG5FBdaGA5THPFj4M,5948
+nshtrainer/callbacks/model_checkpoint.py,sha256=MaDkD8Ismcj8u6l2flCFlqJR3-k1Tc4xzhxNWNux4n0,6556
 nshtrainer/callbacks/norm_logging.py,sha256=T2psu8mYsw9iahPKT6aUPjkGrZ4TIzm6_UUUmE09GJs,6274
 nshtrainer/callbacks/on_exception_checkpoint.py,sha256=x42BYZ2ejf2rhqPLCmT5nyWKhA9qBEosiV8ZNhhZ6lI,3355
 nshtrainer/callbacks/print_table.py,sha256=_FdAHhqylWGk4Z0c2FrLFeiMA4jhfA_beZRK_BHpzmE,2837
@@ -34,7 +35,7 @@ nshtrainer/ll/config.py,sha256=fKumJf42HY2FITX1QUM1OTXkYD6U2np2ciyd4PFRPZ8,145
 nshtrainer/ll/data.py,sha256=zRG0FRje-jtSHximVzkHIHzpwsyQxpHCoACFihNKLPM,44
 nshtrainer/ll/log.py,sha256=d4BB3TyM8imK65EXOiOeUTF0zFM1ropbe7Vq3DeB0xU,140
 nshtrainer/ll/lr_scheduler.py,sha256=7xjhN6L69BCUzFhcy33NtMtPuCzHiB611zVWFg92lQ0,52
-nshtrainer/ll/model.py,sha256=6I9gQjEFT2Veer-UmcPy05Pt3mnvaxYX1b3sOdaF96A,45
+nshtrainer/ll/model.py,sha256=cxFQfFc-2mAYBGwDpP8m5tjQBs7M47cZ6JoPXksPaoI,473
 nshtrainer/ll/nn.py,sha256=8qiRDFwojIxkB7-LtNWk4mLL2tJbaskHYofDsOIHiNg,42
 nshtrainer/ll/optimizer.py,sha256=3T-VZtT73jVvwCNJGDjgGEbzs-1LFTzMQH-SB_58mSo,49
 nshtrainer/ll/runner.py,sha256=B0m5VEhNKIjF1aFmqPkonkQxDoRL2jeHZGsV3zwhSVE,117
@@ -49,10 +50,9 @@ nshtrainer/lr_scheduler/linear_warmup_cosine.py,sha256=mn6cyizyI_stkXtg6zxIEGF9b
 nshtrainer/lr_scheduler/reduce_lr_on_plateau.py,sha256=h76oTHYpMxauV_l6lviya5DW-WKArwxxf7ZQizhmbCw,2782
 nshtrainer/metrics/__init__.py,sha256=ObLIELGguIEcUpRsUkqh1ltrvZii6vglTpJGrPvoy00,50
 nshtrainer/metrics/_config.py,sha256=hWWS4IXENRyH3RmJ7z1Wx1n3Lt1sNMlGOrcU6PW15o0,1104
-nshtrainer/model/__init__.py,sha256=TbexTxiE20WHYg5q3L88Hysk4LlHeKk_isv33aSBREA,1918
-nshtrainer/model/_environment.py,sha256=oTtecQeF5oY2RV7UkkSLnzDy3clz4AUkf9oocD6-e54,23115
-nshtrainer/model/base.py,sha256=WtCj0-nLWeW04Tu2TWVjIq0D-jW_kMN2hg--4VWVnvE,17505
-nshtrainer/model/config.py,sha256=orGBrp8TXnHksfAzXxNJVDdo0X_iIn_nda6BZDS9N70,53349
+nshtrainer/model/__init__.py,sha256=NpvyQHmGaHB8xdraHmm8l7kDHLmvJSgBNQKkfYqtgyI,1454
+nshtrainer/model/base.py,sha256=AXRfEsFAT0Ln7zjYVPU5NgtHS_c8FZM-M4pyLamO7OA,17516
+nshtrainer/model/config.py,sha256=65UDzt3ZZFUQaHMlK7f9wzwyGH3cDyHGtjZ2eOjHvVo,53360
 nshtrainer/model/modules/callback.py,sha256=K0-cyEtBcQhI7Q2e-AGTE8T-GghUPY9DYmneU6ULV6g,6401
 nshtrainer/model/modules/debug.py,sha256=Yy7XEdPou9BkCsD5hJchwJGmCVGrfUru5g9VjPM4uAw,1120
 nshtrainer/model/modules/distributed.py,sha256=ABpR9d-3uBS_fivfy_WYW-dExW6vp5BPaoPQnOudHng,1725
@@ -73,11 +73,12 @@ nshtrainer/trainer/_runtime_callback.py,sha256=sd2cUdRJG-UCdQr9ruZvEYpNGNF1t2W2f
 nshtrainer/trainer/checkpoint_connector.py,sha256=xoqI2dcPnlNFPPLVIU6dBOvRPC9PtfX5qu__xV1lx0Y,2124
 nshtrainer/trainer/signal_connector.py,sha256=llwc8pdKAWxREFpjdi14Bpy8rGVMEJsmJx_s2p4gI8E,10689
 nshtrainer/trainer/trainer.py,sha256=tFyzIsF8c-FABTH6wwDOR9y8kydVJqeVO7PDNFMvhSU,16950
+nshtrainer/util/_environment_info.py,sha256=yPtAbgjCY4tkvh5wp9sjNsF0Z45TYwzEAM_N2_b5BbY,23123
 nshtrainer/util/environment.py,sha256=AeW_kLl-N70wmb6L_JLz1wRj0kA70xs6RCmc9iUqczE,4159
 nshtrainer/util/seed.py,sha256=Or2wMPsnQxfnZ2xfBiyMcHFIUt3tGTNeMMyOEanCkqs,280
 nshtrainer/util/slurm.py,sha256=rofIU26z3SdL79SF45tNez6juou1cyDLz07oXEZb9Hg,1566
 nshtrainer/util/typed.py,sha256=NGuDkDzFlc1fAoaXjOFZVbmj0mRFjsQi1E_hPa7Bn5U,128
 nshtrainer/util/typing_utils.py,sha256=8ptjSSLZxlmy4FY6lzzkoGoF5fGNClo8-B_c0XHQaNU,385
-nshtrainer-0.10.10.dist-info/METADATA,sha256=AQipVj-dOXT3cPwzbfvg6u3KdgLocc1BeaSLjndnL1A,696
-nshtrainer-0.10.10.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-nshtrainer-0.10.10.dist-info/RECORD,,
+nshtrainer-0.10.12.dist-info/METADATA,sha256=RDwRSo6cvM5H1OgeCc3C5ZglfwjhAQ0I1nlw4xfLpUw,696
+nshtrainer-0.10.12.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+nshtrainer-0.10.12.dist-info/RECORD,,